Shubhankar Mittal

Designed and integrated LLM-driven backend services (Python, FastAPI) for BFSI workflow orchestration including intent classification, RAG-based retrieval, and rule-based validation. Architected an agent orchestration framework routing requests across specialized LLM agents using confidence thresholds, fallbacks, policy checks, and improved response reliability. Optimized asynchronous pipelines, retrieval logic, and prompt structure, achieving ~60% 95th percentile latency reduction and ~40% error reduction in pre-production evaluation. Established structured evaluation and logging pipelines across recurring internal test cycles to analyze model behavior and iteratively improve response consistency.

Engineered Python-based analytics and risk mitigation frameworks used across multiple BFSI client engagements, automating compliance workflows and reducing quarterly analysis cycle time by ~20 days. Modeled scalable batch data-processing pipelines over 10M+ transactional records to support segmentation, forecasting, and ML-driven decision making in regulated environments. Led hands-on development and code reviews, improving data quality checks and reducing downstream rework by ~35%. Conducted tier-based vendor assessments for AI/GenAI platforms assessing capability, cost, security, and regulatory fit to guide technology selection.

Deployed Python-based automation and analytics components across multiple compliance monitoring processes, reducing identified conduct-risk incidents by ~17%. Developed data ingestion, feature engineering, and analytics pipelines; partnered with engineering teams to productize analytics logic, improving turnaround time by ~40%.